jax
fix: reduce DPX workers_per_gpu 2→1 to avoid HBM exhaustion
#890

Open

fix: reduce DPX workers_per_gpu 2→1 to avoid HBM exhaustion #890

mkunredd wants to merge 3 commits into magaonka/dpx_experiment from fix/dpx-workers-1-per-partition

fix: reduce DPX workers_per_gpu 2→1 to avoid HBM exhaustion

fbd01185

fix: strip \r from amd-smi output + reduce DPX workers to 1 per parti…

4475474a

fix: robust amd-smi DPX detection + workers_per_gpu=1 for DPX

95345ac1

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Milestone

No milestone